Sage Journals: Discover world-class research

Abstract

BACKGROUND:

Cervical lymph node metastases (LNM) in papillary thyroid carcinomas (PTCs) are common and develop in approximately 30–80% of PTCs. The presence of cervical LNM significantly increases the rate of locoregional recurrence in PTCs.

OBJECTIVE:

To search for predictive gene signatures for nodal metastasis in PTCs.

METHODS:

We used unsupervised clustering with unbiased manner to compare molecular profiles between PTCs with nodal metastasis and PTCs without nodal metastasis using mRNA-seq of TCGA data. Using gene ontology (GO) and logistic regression test, we generated 12-predictive genes for nodal metastasis in PTCs.

RESULTS:

Unsupervised clustering of mRNA-seq (training set, N $=$ 158) revealed that PTCs with nodal metastasis showed different gene expression patterns compared to PTCs without nodal metastasis. We generated 12 predictive genes and these gene signatures showed consistency for predicting nodal metastasis when we applied them to a validation set (N $=$ 80). Based on multivariate analysis, these 12 predictive gene signatures showed more significant odds ratio compared to other variables.

CONCLUSIONS:

These 12 gene signatures could be used to predict the chance of nodal metastasis in PTCs in preoperative evaluation using fine needle aspiration biopsy (FNAB) so that appropriate plan such as central neck dissection could be made.

Keywords

Gene expression PTC node metastasis prediction gene signature

1. Introduction

Papillary thyroid carcinoma (PTC) is the most common type of thyroid cancer. Approximately about 80% of all thyroid malignancies are PTCs [1]. Cervical lymph node metastases (LNM) in PTCs are common and develop in approximately 30–80% of PTCs [2, 3, 4]. The impact of neck node metastases on the prognosis of patients with PTCs is controversial [5, 6, 7]. Although many studies have reported that cervical LNM has no major impact on survival of low-risk patients [8], recently there has been growing recognition that cervical LNM can adversely affect survival, particularly in older patients with larger tumors and extrathyroidal extension (ETE) [9, 10, 11]. In contrast to its impact on cancer prognosis, the impact of cervical LNM on cancer recurrence is undisputed. The presence of LNM significantly increases the rate of locoregional recurrence [12, 13]. Since 60% to 75% of all cancer recurrence in the neck occur in lymph nodes, detection of cervical LNM at initial preoperative evaluation is very important to reduce reoperation rates [14].

Ultrasonography has been reported to be the most sensitive method to detect metastatic lymph nodes in patients with PTC [15, 16, 17]. This technique is capable of detecting metastatic lymph nodes as small as 2 to 3 mm in diameter [17]. Therefore, preoperative ultrasonographic staging of the neck might improve surgical management and reduce recurrence. However, 50% to 90% of subclinical micrometastases could not be detected by preoperative ultrasonographic evaluation [18, 19, 20].

Several studies have reported that clinicopathological variables could predict cervical LNM in PTCs. Koo et al. have reported that PTCs with maximal diameter of greater than 1-cm is associated with a high rate of ipsilateral central neck LNM [21]. Kupferman et al. have reported that multifocal cancer within the thyroid is associated with a high rate of ipsilateral lateral neck LNM [22]. However, whether gene signatures could be used to predict nodal metastasis in PTCs is currently unknown.

Therefore, the objective of this study was to compare gene expression profiles between PTCs with nodal metastasis and PTCs without nodal metastasis using mRNA-seq of TCGA data and search for gene signatures that could predict nodal metastasis in PTCs using multiplatform genomic analysis.

2. Methods

2.1 Genomic and clinical data sets

All genomic data of papillary thyroid carcinoma from TCGA project were obtained from TCGA data portal (https://tcga-data.nci.nih.gov) and cancer browser (https://genome-cancer.ucsc.edu). Gene expression data from mRNA-seq (training cohort, N $=$ 158; validation cohort, N $=$ 80) and clinical information for both cohorts (N $=$ 158 and N $=$ 80) were included in analyses. Clinical data included age, gender, unifocal/multifocal, extrathyroid extension, and TNM stage (Supplementary Datas 1 and 2).

2.2 Analysis of gene expression data and unsupervised clustering

BRB-ArrayTools software program (http://linus.nci.nih.gov/BRB-ArrayTools.html) was used to analyze gene expression data [23]. After gene expression data were gene-median centered, gene variability was computed using median absolute deviation. A total of 6589 most variable genes were selected. Heatmap was generated using Cluster and TreeView software programs [24]. Other statistical analyses were performed in R language (http://www.r-project.org).

2.3 Selection of specific gene signature in each cluster

To select genes differentially expressed between PTCs with nodal metastasis and PTCs without nodal metastasis, we applied stringent cutoff of $P<$ 0.001 (Student’s $t$ -test) and 1.5-fold difference.

2.4 Significant canonical signaling pathways enriched in each subgroup

Pathway analysis was carried by using Ingenuity Pathways Analysis (IPA, Ingenuity, Redwood City, CA, USA). Genes from the dataset that were associated with a canonical pathway in Ingenuity Pathways Knowledge Base were considered for analysis. The significance of the association between differently expressed genes and the canonical pathway was measured using Fischer’s exact test ( $P<$ 0.001).

Figure 1.

Unsupervised clustering heat-map showing two distinct molecular patterns of gene expression between PTCs with nodal metastasis and PTCs without nodal metastasis. A hierarchical clustering of gene expression data from 158 PTCs cases in TCGA data was performed. Genes with expression levels at least 2-fold different in at least 15 cases relative to median value across cases were selected for hierarchical clustering analysis. Data are given in matrix format, in which rows represent individual genes while columns represent each patient. Each cell in the matrix represents the expression level of gene feature in an individual pattern. Color red or green in cell reflects relative high or low expression levels, respectively, as indicated in the scale bar.

Table 1

Altered canonical pathways in node positive PTCs

Altered Pathways	$-$ log ( $p$ -value)	Ratio	Molecules
Dendritic cell maturation	1.53E00	2.63E-02	COL1A1, CD1A, COL10A1, CSF2, CD1C
Colorectal cancer – metastasis signaling	1.13E00	2.02E-02	MMP7, WNT10A, MMP13, WNT4, ADCY8
GABA receptor signaling	2.5E00	5.97E-02	KCNN4, GABRB3, GABRG1, ADCY8
LXR/RXR activation	1.62E00	3.31E-02	IL1RL1, CD36, SERPINA1, CETP
Wnt/ $\beta$ -catenin signaling	1.72E00	2.96E-02	SFRP4, MMP7, SFRP2, WNT10A, WNT4

PTCs: Papillary thyroid carcinomas.

2.5 Selection of specific gene signatures to predict nodal metastasis in PTCs

To select gene signatures that could predict nodal metastasis in PTCs, we used 200 differently expressed genes (DEG) between PTCs with nodal metastasis and PTCs without nodal metastasis (Supplementary Data 3). We tried to pick up representative genes for each biologic pathway. For this analysis, we used Gene Ontology Consortium online tools (http://geneontology.org/) and divided 200 DEG into several subgroups which belonged to the same biologic category. Among these genes within the same biologic category, we used multivariate logistic regression test to select one or more drive genes which are representative of other genes in same subgroup, and selected genes with statistically significant $p$ value ( $p<$ 0.05).

Figure 2.

Heat-map showing each subtype-specific gene expression pattern. Data given in the matrix represent the expression of a gene feature in an individual patient. We selected nodal metastasis-specific 200 genes. Heat-map is ordered by ratio. Color red or green in cell reflects relative high or low expression levels, respectively.

2.6 Prediction with Bayesian compound covariate predictor (BCCP)

A total of 12 predictive gene signatures were used to evaluate their robustness in predicting nodal metastasis in a validation cohort (N $=$ 80, Supplementary Data 2). Briefly, expression data for 12 predictive genes in the training set were combined to form a classifier according to BCCP [25]. The BCCP classifier estimated the likelihood that an individual patient was included in one of two subgroups. According to a BCCP, $p$ -value of 0.6 which was optimized by results of the current analysis of receiver operating characteristic (ROC) was used to predict separate validation cohorts. The cutoff was set by maximal point of sum of sensitivity and specificity [26].

2.7 Statistical analysis

The association of each group with each clinical variable was evaluated using Chi-square test and Fisher’s exact test. $P$ value of less than 0.05 was considered statistically significant. All statistical analyses were conducted in R language environment (http://www.r-project.org).

Figure 3.

(A) Heat-map showing definitely distinct gene expression pattern between PTCs with node positive (left, red-bar) and PTCs with node negative (right, blue-bar) in training cohort. (B) ROC curve showing the predictive value of 12 predictive genes in training cohort (N $=$ 158).

Figure 4.

(A) Heat-map showing distinct gene expression pattern between PTCs with node positive (left, red-bar) and PTCs with node negative (right, blue-bar) in validation cohort. (B) ROC curve showing the predictive value of 12 predictive genes in validation cohort (N $=$ 80).

3. Results

3.1 Unsupervised clustering of training cohort (N $=$ 158)

To compare gene expression patterns between PTCs with nodal metastasis and PTCs without nodal metastasis (N $=$ 158), we performed hierarchical clustering in an unsupervised and unbiased manner using cluster 3.0. When we used heat-map to compare gene expression profiles, we found that PTCs with nodal metastasis showed different gene expression patterns compared to PTCs without nodal metastasis (Fig. 1).

3.2 Selection of differently expressed genes in each group (PTCs with nodal metastasis vs. PTCs without nodal metastasis)

Next, we tried to select genes that were differentially expressed between PTCs with nodal metastasis and PTCs without nodal metastasis. We applied stringent cutoff of $P<$ 0.001 (Student’s t-test) and 1.5-fold difference and identified 200 differently expressed genes (Fig. 2 and Supplementary Data 3). In Fig. 2, data given in the matrix represent expression level of a gene feature in an individual patient. Color red or green in cell reflects relative high or low expression level. Upper 100 genes were highly expressed in PTCs with nodal metastasis while lower 100 genes were highly expressed in PTCs without nodal metastasis (Fig. 2).

3.3 Significant canonical signaling pathways enriched in PTCs with nodal metastasis

Pathway analysis revealed that PTCs with nodal metastasis group showed different canonical signaling pathways (Table 1 and Supplementary Data 4). PTCs with nodal metastasis showed enriched canonical pathways such as GABA receptor signaling pathway, LXR/ RXR activation pathway, and Wnt/ $\beta$ -catenin signaling. $P$ value and ratio of each pathway are shown in Supplementary Data 4.

3.4 Selection of specific gene signatures to predict nodal metastasis in PTCs

We used Gene Ontology Consortium online tools and logistic regression test to select representative genes for each biologic pathway. Finally, a total of 12 genes (BCC8, CHI3L1, CLCNKA, FAM155B, GABRG1, LUM, MRO, MT1G, MT1H, SELV, SLC4A4, and TMEM92) were selected (Supplementary Data 5).

3.5 Prediction with BCCP for validation cohort

These 12 predictive gene signatures (Supplementary Data 5) were used to predict nodal metastasis in PTCs using a validation cohort (N $=$ 80). Heatmap and ROC curve of these 12 predictive genes in training cohort (N $=$ 158) and validation cohort (N $=$ 80) are shown in Figs 3 and 4, respectively.

Table 2
Univariate analysis of clinical variables in validation cohort (N $=$ 80)

		Node ( $+$ )	Node ( $-$ )	$p$ value
Age (years old)	Mean	50.1	50.5	$p=$ 0.92
Tumor depth (cm)	Mean	1.7	1.8	$p=$ 0.45
Gender	Male	17	10	$p=$ 0.72
	Female	28	22
Multifocality	Unifocal	16	20	$p=$ 0.03*
	Multifocal	29	12
Extrathyroidal	ETE( $+$ )	23	6	$p=$ 0.008*
extension (ETE)	ETE( $-$ )	22	26
12_gene_	Positive	35	6	$p=$ 1.03
signature	Negative	10	26	$\times$ 10 ${}^{-6}$ *.

*means statistically significant $p$ value.

Figure 5.

Plot for odds ratios of each variable in validation cohort. Logistic regression with univariate analysis revealed that ETE, multifocality, and 12-gene signatures were significantly correlated with nodal metastasis. Multivariate analysis revealed that these 12 predictive gene signatures had significantly higher odds ratio than other variables.

3.6 Logistic regression analysis and odds ratio of each clinical variable

Results of logistic regression with univariate analysis revealed that ETE, multifocality, and these 12 gene signatures had statistically significant correlations with nodal metastasis. Multivariate analysis revealed that these 12 predictive gene signatures showed more significant odds ratio compared to other variables (Fig. 5 and Table 2).

4. Discussion

Cervical LNM is common (30% to 80%) in patients with PTC [1, 2]. Although the effect of nodal metastasis on survival is marginal, the presence of LNM significantly increases the rate of locoregional recurrence [2, 3, 4, 5, 6]. For this reason, preoperative detection of metastatic node is important for selecting treatment options.

Several reports have described predictive factors for central nodal metastasis in patients with different thyroid cancers. Extracapsular spread (ECS), lateral LNM, and tumor size exceeding 5 mm are significantly associated with central compartment nodal metastasis in papillary thyroid carcinoma [1]. A maximal diameter exceeding 1 cm has been associated with a high rate of ipsilateral central LNM [11]. However, there has been no report about gene signatures for predicting cervical nodal metastasis in PTCs.

In this study, we generated 12-gene signatures and found that they could predict nodal metastasis of PTC. Their predictive value was evaluated in another cohort and odd ratios of these gene signatures were compared to those of other clinical variables. Odd ratios of these gene signatures for predicting nodal metastasis of PTC was found to be much higher than those of other clinical variables such as ETE or multifocality. With these gene signatures, we can predict the chance of nodal metastasis in PTCs in preoperative evaluation using FNAB to make appropriate plan including central neck dissection.

Recently, large-scale multiplatform analysis of genomic data including gene expression, copy number alteration, somatic mutation, and methylation data has been performed, making it possible to compare two or more tumors with different clinical behavior at molecular level. These genomic studies have revealed that same-tissue origin cancers could be divided into multiple molecular subtypes [27]. These sub-classifications of cancer type make it possible to not only predict clinical behavior, but also suggest appropriate targeted therapy considering specific pathway involved in cancer development and progress [27]. Using gene expression data of PTCs in TCGA data, we suggested 12 gene-signatures in this study to predict nodal metastasis in PTCs. When we used heat-map and ROC curve to evaluate the predictive value of these 12 gene signatures, heat-map in training cohort (N $=$ 158) showed very distinct distribution pattern of gene expression and the area under curve (AUC) for training cohort was 99.6% (99.0%–100.0%, 95% confidence interval) (Fig. 3). However, when we applied these predictive gene signatures to validation cohort (N $=$ 80), heat-map showed some mixed expression patterns with AUC of 89.0% (80.8%–97.2%, 95% confidence interval) (Fig. 4). Since we selected these 12 predictive gene signatures from the training cohort, AUC of these 12 genes for the training cohort was found to be higher than that for the validation cohort. Difference in case number between training cohort and validation cohort might have also contributed to such discrepancy in AUC. However, when we compared odds ratio of gene signatures with other clinical variables for the validation cohort, odds ratios of gene signatures were much higher than those of other variables such as ETE or multifocality. Based on logistic regression with univariate analysis, ETE, multifocality, and these 12 gene signatures showed statistically significant correlations with nodal metastasis. Results of multivariate analysis revealed that these 12 predictive gene signatures showed significantly higher odds ratios compared to other variables (Fig. 5).

Cost for microarray analysis of the whole genome has sharply decreased as the price of genome analysis was $100 million in 2001 but now (2017) is less than $500. Development of multiplatform genetic techniques made it possible to reduce the price of genomic analysis. In the near future, the price of genomic analysis might be cheaper than ultrasound or computerized tomography (CT). This is why we believe the study of genomic analysis is absolutely needed even with little higher cost.

Prophylactic central neck dissection of PTCs may decrease locoregional recurrence cause by occult nodal metastasis [28]. Opponents of prophylactic neck dissection contend that microscopic nodal metastasis can be treated with radioactive iodine ablation therapy after total thyroidectomy without neck dissection [29]. In addition, operation-related morbidity such as recurrent nerve injury and parathyroid injury might be caused by prophylactic neck dissection [29]. Thus, it is most likely that prophylactic neck dissection is beneficial at the time of initial surgery for selected high-risk patients. Determining which patients are at high-risk before surgery remains difficult and important. In this study, we found that 12 gene-signatures could predict nodal metastasis in PTCs. Therefore, these gene signatures could be used in preoperative FNAB procedure. Using biopsy sample, we can get array data and exact information about nodal metastasis. Such prediction of nodal metastasis will be really helpful in planning surgical extent. Sometimes, preoperative ultrasound or other radiologic evaluation fail to detect small metastatic lymph nodes of PTC. In this case, if these 12 predictive gene signatures showed a high risk pattern, more comprehensive preoperative ultrasound or CT evaluation should be needed to detect occult metastatic nodes. Also, as for the surgical extent of thyroid cancer surgery, we could perform elective central neck dissection for those high-risk patients.

Zhao et al. have suggested that methylation of DACT2 promotes PTCs metastasis by activating Wnt/ $\beta$ -catenin signaling pathway [30]. They reported that expression levels of beta-catenin, c-myc, cyclin D1, and MMP-9 were increased after knockdown of DACT2in cell lines [30]. In this study, PTCs with nodal metastasis were enriched in canonical pathways such as GABA receptor signaling pathway, LXR/RXR activation pathway, and Wnt/ $\beta$ -catenin signaling (Table 1). Functional studies for these pathways are needed in the future.

This is the first study that uses genomic data to suggest gene signatures for predicting nodal metastasis of PTCs and compares these gene signatures to other clinical variables that might have prognostic effect on nodal metastasis. Our results could be used to predict nodal metastasis in PTCs and plan surgical extent such as elective central neck dissection even if there is no evidence of nodal metastasis in preoperative radiologic evaluation. The limitation of this study is its retrospective nature. A prospective and functional study is needed in the future to obtain precise information about the role of these 12 gene signatures in predicting nodal metastasis in PTCs.

In conclusion, using gene expression data and bioinformatic analysis, we generated 12 gene signatures that could predict nodal metastasis of PTC. With these gene signatures, we can predict the chance of nodal metastasis in PTCs in preoperative evaluation staging using FNAB. This can benefit appropriate planning, including central neck dissection.

Footnotes

Acknowledgments

We appreciated patients and their families who generously donated their tissues to TCGA, as well as the members of TCGA who collected and disclosed valuable data. This research was supported by Hallym University Research Fund 2017 (HURF-2017-82).

Conflict of interest

The authors declare no competing financial interests and no other conflicts of interest.

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-170784.

References

Sherman

S.I.

, Thyroid carcinoma, Lancet 361 (2003), 501–511.

Shaha

A.R.

, Management of the neck in thyroid cancer, Otolaryngol Clin North Am 31 (1998), 823–831.

Shaha

A.R.

Shah

J.P.

and Loree

T.R.

, Patterns of nodal and distant metastasis based on histologic varieties in differentiated carcinoma of the thyroid, Am J Surg 172 (1996), 692–694.

McGregor

G.I.

Luoma

and Jackson

S.M.

, Lymph node metastases from well-differentiated thyroid cancer: A clinical review, Am J Surg 149 (1985), 610–612.

Bacourt

Asselain

Savoie

J.C.

D’Hubert

Massin

J.P.

Doucet

Leger

and Garnier

, Multifactorial study of prognostic factors in differentiated thyroid carcinoma and a re-evaluation of the importance of age, Br J Surg 73 (1986), 274–277.

Cunningham

M.P.

Duda

R.B.

Recant

Chmiel

J.S.

Sylvester

J.A.

and Fremgen

, Survival discriminants for differentiated thyroid cancer, Am J Surg 160 (1990), 344–347.

Hannequin

Liehn

J.C.

and Delisle

M.J.

, Multifactorial analysis of survival in thyroid cancer: Pitfalls of applying the results of published studies to another population, Cancer 58 (1986), 1749–1755.

Shah

J.P.

Loree

T.R.

Dharker

Strong

E.W.

Begg

and Vlamis

, Prognostic factors in differentiated carcinoma of the thyroid gland, Am J Surg 164 (1992), 658–661.

Lundgren

C.I.

Hall

Dickman

P.W.

and Zedenius

, Clinically significant prognostic factors for differentiated thyroid carcinoma: A population-based, nested case-control study, Cancer 106 (2006), 524–531.

10.

Bhattacharyya

, Surgical treatment of cervical nodal metastases in patients with papillary thyroid carcinoma, Arch Otolaryngol Head Neck Surg 129 (2003), 1101–1104.

11.

Simon

Goretzki

P.E.

Witte

and Roher

H.D.

, Incidence of regional recurrence guiding radicality in differentiated thyroid carcinoma, World J Surg 20 (1996), 860-6; discussion 866.

12.

Wada

Duh

Q.Y.

Sugino

Iwasaki

Kameyama

Mimura

Ito

Takami

and Takanashi

, Lymph node metastasis from 259 papillary thyroid microcarcinomas: Frequency, pattern of occurrence and recurrence, and optimal strategy for neck dissection, Ann Surg 237 (2003), 399–407.

13.

Shaha

A.R.

Shah

J.P.

and Loree

T.R.

, Patterns of failure in differentiated carcinoma of the thyroid based on risk groups, Head Neck 20 (1998), 26–30.

14.

Watkinson

J.C.

Franklyn

J.A.

and Olliff

J.F.

, Detection and surgical treatment of cervical lymph nodes in differentiated thyroid cancer, Thyroid 16 (2006), 187–194.

15.

Hwang

H.S.

and Orloff

L.A.

, Efficacy of preoperative neck ultrasound in the detection of cervical lymph node metastasis from thyroid cancer, Laryngoscope 121 (2011), 487–491.

16.

Roh

J.L.

Park

J.Y.

Kim

J.M.

and Song

C.J.

, Use of pre-operative ultrasonography as guidance for neck dissection in patients with papillary thyroid carcinoma, J Surg Oncol 99 (2009), 28–31.

17.

Antonelli

Miccoli

Ferdeghini

Di Coscio

Alberti

Iacconi

Baldi

Fallahi

and Baschieri

, Role of neck ultrasonography in the follow-up of patients operated on for thyroid cancer, Thyroid 5 (1995), 25–28.

18.

Grebe

S.K.

and Hay

I.D.

, Thyroid cancer nodal metastases: Biologic significance and therapeutic considerations, Surg Oncol Clin N Am 5 (1996), 43–63.

19.

Scheumann

G.F.

Gimm

Wegener

Hundeshagen

and Dralle

, Prognostic significance and surgical management of locoregional lymph node metastases in papillary thyroid cancer, World J Surg 18 (1994), 559-67; discussion 567–568.

20.

Wang

T.S.

Dubner

Sznyter

L.A.

and Heller

K.S.

, Incidence of metastatic well-differentiated thyroid cancer in cervical lymph nodes, Arch Otolaryngol Head Neck Surg 130 (2004), 110–113.

21.

Koo

B.S.

Choi

E.C.

Yoon

Y.H.

Kim

D.H.

Kim

E.H.

and Lim

Y.C.

, Predictive factors for ipsilateral or contralateral central lymph node metastasis in unilateral papillary thyroid carcinoma, Ann Surg 249 (2009), 840–844.

22.

Kupferman

M.E.

Weinstock

Y.E.

Santillan

A.A.

Mishra

Roberts

Clayman

G.L.

and Weber

R.S.

, Predictors of level V metastasis in well-differentiated thyroid cancer, Head Neck 30 (2008), 1469–1474.

23.

Simon

Lam

M.C.

Ngan

Menenzes

and Zhao

, Analysis of gene expression data using BRB-Array Tools, Cancer Inform 4 (2007), 11–7.

24.

Eisen

M.B.

Spellman

P.T.

Brown

P.O.

and Botstein

, Cluster analysis and display of genome-wide expression patterns, Proc Natl Acad Sci USA 95 (1998), 14863–14868.

25.

Ramaswamy

Tamayo

Rifkin

Mukherjee

Yeang

C.H.

Angelo

Ladd

Reich

Latulippe

Mesirov

J.P.

Poggio

Gerald

Loda

Lander

E.S.

and Golub

T.R.

, Multiclass cancer diagnosis using tumor gene expression signatures, Proc Natl Acad Sci USA 98 (2001), 15149–15154.

26.

DeLong

E.R.

DeLong

D.M.

and Clarke-Pearson

D.L.

, Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach, Biometrics 44 (1988), 837–845.

27.

Hoadley

K.A.

Yau

Wolf

D.M.

Cherniack

A.D.

Tamborero

Leiserson

M.D.

Niu

McLellan

M.D.

Uzunangelov

Zhang

Kandoth

Akbani

Shen

Omberg

Chu

Margolin

A.A.

Van’t Veer

L.J.

Lopez-Bigas

Laird

P.W.

Raphael

B.J.

Ding

Robertson

A.G.

Byers

L.A.

Mills

G.B.

Weinstein

J.N.

Van Waes

Chen

Collisson

E.A.

Cancer Genome Atlas Research

Benz

C.C.

Perou

C.M.

and Stuart

J.M.

, Multiplatform analysis of 12 cancer types reveals molecular classification within and across tissues of origin, Cell 158 (2014), 929–944.

28.

White

M.L.

Gauger

P.G.

and Doherty

G.M.

, Central lymph node dissection in differentiated thyroid cancer, World J Surg 31 (2007), 895–904.

29.

Shaha

A.R.

, Implications of prognostic factors and risk groups in the management of differentiated thyroid cancer, Laryngoscope 114 (2004), 393–402.

30.

Zhao

Herman

J.G.

Brock

M.V.

Sheng

Zhang

Liu

and Guo

, Methylation of DACT2 promotes papillary thyroid cancer metastasis by activating Wnt signaling, PLoS One 9 (2014), e112336.

Predictive gene signatures of nodal metastasis in papillary thyroid carcinoma

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Methods

2.1 Genomic and clinical data sets

2.2 Analysis of gene expression data and unsupervised clustering

2.3 Selection of specific gene signature in each cluster

2.4 Significant canonical signaling pathways enriched in each subgroup

2.7 Statistical analysis

3.1 Unsupervised clustering of training cohort (N = 158)

3.2 Selection of differently expressed genes in each group (PTCs with nodal metastasis vs. PTCs without nodal metastasis)

3.3 Significant canonical signaling pathways enriched in PTCs with nodal metastasis

3.4 Selection of specific gene signatures to predict nodal metastasis in PTCs

3.5 Prediction with BCCP for validation cohort

Table 2 Univariate analysis of clinical variables in validation cohort (N = 80)

4. Discussion

Footnotes

Acknowledgments

Conflict of interest

Supplementary data

References

3.1 Unsupervised clustering of training cohort (N $=$ 158)

Table 2
Univariate analysis of clinical variables in validation cohort (N $=$ 80)